News Attention Degree Prediction Based On Bayesian Network

With the development of TouTiao,Netease news and other major news platforms.Based on the Internet,a huge amount of Web news has been produced,and the scale of it has been increasing rapidly.Web news has become the mainstream channel for information acquisition,because it is real-time,interactive and convenient features.While bringing convenience,the increasing number of Web news has brought the problem of information explosion.It is difficult for users to filter out the news they are interested in from massive news.Due to the excessive number of news,it is difficult for news platforms to publish or push news in a targeted manner.Using the characteristics that people are often more likely to be interested with high attention degree news,news attention degree prediction can provide solutions to these problems.First,the news platform managers can select the news with high attention degree to push it to users,based on the predicted news attention degree.The news platform managers can also use the predicted news attention degree to improve the layout of news website.Predicting news attention degree can provide a strategy for online advertising(e.g.choosing to put an advertisement in it after predicting that a news will have a high attention degree,so as to increasing the click-through rate of the advertisement).It also helps to improve the effectiveness of the information and avoid users being disturbed be uninterested information.At present,researchers have proposed many methods for predicting news attention degree,such as classification method or regression method,neural networks and support vector machine.Among them,classification or regression is the most widely used method.The idea of classification or regression method is to represent news as a set of features based on the influence of news attention degree,and use these features to train models to predict news attention degree.However,these methods do not take into account the interdependence of features related to news attention degree.As an effective tool for uncertain knowledge representation and inferencing,Bayesian Network can be used to model the dependencies among these features and use it for inferencing.This thesis builds a model based on Bayesian Network to predict news attention degree.Specifically speaking,the contents of the thesis are as follows:(1)In the process of extracting features for building model,aiming at the problem that there are too many keywords to measure the influence of each keyword on news attention degree independently.We use different levels of keyword sets to represent the influence of keywords on news attention degree,and this method transforms the research problem of the influence of keywords on news attention degree into the study of the influence of news keyword level on news attention degree.(2)In the process of building NABN model for predicting news attention degree,the structure constructed by BIC score combined with hill-climb search method has unreasonable dependence relationship.We propose a structure learning method based on constraint and scoring-search to constrain the structure learning to ensure that the structural dependence relationship is reasonable.(3)We experiment with more than 17000 pieces of news data crawled by TouTiao,tests the effectiveness and feasibility of this method.Designing and implementing the corresponding prototype system.